Research Of Acupuncture Point Signal Characteristics Analysis And Classification Algorithms

Meridians theory is one of the cores in the basic theory of Traditional Chinese Medicine (TCM). It plays a decisive role in the guidance of the TCM clinical practice. Meridians are the main channels and pathways between the body and organs, the physical organisms and the external environment. The physiological role of the meridian is manifested through its dynamic transmission of information and material. Therefore, the research on the meridian characteristics, especially within the information flow, physiological regulation mechanism, and the relation between the meridians and physiological and pathological states, has important implications to meridians clinical diagnosis, prevention and treatment, and of great reference to further research on the meridian substantiality. In this dissertation, the electrical signal of acupuncture point is preprocessed by a well designed Kalman Filter firstly. Then with the introduction of the time-frequency domain analysis, high-order spectrum analysis and the nonlinear dynamic systems analysis, the characteristic parameters of the acupoint signals are extracted and analyzed. Finally, the classification algorithms between the acupoint signals and the non-acupoint signals had been discussed. The main contributions of the dissertation are summarized as follows.A filter used for denoising the acupoint electrical signals is designed. On the basis of the analysis of the factors and difficulties encountered in the acupoint electrical signal generation mechanism and the acquisition, the Kalman filter was chosen to process the signal. Also, considering the weak robustness of the Kalman filter, the stabilization treatment is carried out in uncertainty case of the existence probability model. The experiment results show that the proposed robust Kalman filtering method has better filtering effect both on acupoint signals and non-acupoint signals.The characteristic analysis on acupoint signals was carried out from both the time-domain and frequency-domain. The theory and implementation of the short-time Fourier transform are firstly discussed. Then, the wavelet transform and Wigner distribution analysis are analyzed. Based on the study on static time-frequency analysis, dynamic time-frequency analysis of Gabor transform and wavelet entropy method is introduced to analyze the acupoint signals.Higher order spectral analysis is applied to acupoint signals. A new composite slice spectrum computing method is proposed for the analysis of the characteristics of acupoint signals, where the computing speed and comprehensiveness of the information is taken into account. Time complexity analysis and experiments results show that the proposed method retains information on the horizontal and vertical directions. Furthermore, it reduces the amount of computation, consumes less time, and improves the operating speed. On the basis of high-order spectrum analysis, the Wigner time varying higher order spectrum and its slices are applied for the analysis on acupoint signals. Experiment results show that there exist obvious differences between the healthy human body acupoint signals and the corresponding non-acupuncture point’s signals, acupoint signals between different states of before meals and after meals, acupoint signals between healthy human body and patients of cardiovascular disease. For the signal from low-frequency sub-band of the wavelet decomposition, its high order spectral characteristics of after meals was significantly higher than that of before meals; while the high-frequency sub-band signals show the opposite trend. The high-order spectral features of acupoint signal from patients with cardiovascular disease significantly lower than that of healthy humam bodys.Chaotic Characteristic Analysis of the acupoint signals has been carried out. Firstly, the surrogate-data technique is chosen to detect the nonlinear property of the acupoint electrical signals. The results have shown the nonlinearities of the acupoint signals. Then, the phase spaces are reconstructed for the acupoint signals, the corresponding non-acupoint signals and signals from patients of cardiovascular disease respectively, and then the commonly-used nonlinear dynamics characteristics are analyzed. The experimental results reveal that the acupoint and non-acupoint signals are both chaotic, while the corresponding nonlinear dynamics parameters are different to a certain extent.The classification methods based on optimized neural networks are studied. To deal with the problems with the traditional BP algorithms such as the slow convergence speed, local minimum, less robustness on multivalent classification, a quantum evolution neural networks model is developed in this dissertation for the classification of the acupoint electrical signals. The experimental results have shown that the model gives satisfactory classification ability between the acupoint signals and the non-acupoint signals, acupoint signals between different states of before meals and after meals, acupoint signals between healthy human body and patients of cardiovascular disease.The support vector machine algorithm has also been implemented for the classification of the meridians signals. A particle swarm optimization based support vector machine classification model is proposed to meet the difficulty in the choice of the proper parameter in the support vector machine.The experiment results have shown that the model has exhibited an effective classification performance on acupoint signals and non-acupoint electrical signals, acupoint signals between different states of before meals and after meals, acupoint signals between healthy human body and patients of cardiovascular disease. Meanwhile, it gives a better performance than the neural network classification model.